Organisations that procure a second cloud platform for resilience usually acquire complexity instead. Failover between platforms requires portable architecture, duplicated state, parallel identity and networking, runbooks rehearsed on both platforms, and engineers fluent in both. Platform incidents commonly affect control planes and identity — the layer that duplication must get right.

What separation costs

  • Skills. Dual-platform engineers are scarcer and more expensive than single-platform engineers; teams split across platforms lose depth in both.
  • Tooling and governance. Either two of everything — security posture, cost views, deployment pipelines — or an abstraction layer, which leaks during the failovers it exists to serve.
  • Drift. Two estates on two vendors’ release cadences diverge; divergence is discovered during incidents.
  • Exit terms. Portability evidence, egress positions and contract clauses negotiated while leverage exists — before signing.

When the price is justified

  • Sovereignty and jurisdiction. Obligations that place specific data or workloads on specific platforms.
  • Acquisition. A second platform arrives with a purchase; operating one estate across both is faster and safer than forced consolidation.
  • Latency and locality. Customers or regulators in regions where platforms differ materially in presence.
  • Negotiating leverage at scale. Credible exit is the cheapest discount available to large buyers.

Outside these four cases, the same budget buys more resilience on the platform already in use: regional architecture, rehearsed recovery, and redundancy reviewed against the services that actually matter.

If you proceed

  • One identity fabric across platforms, with its failure modes tested.
  • A written rule for which services use portable architecture and which use platform-native features.
  • One security posture view and one cost view, or an explicit decision to run two estates.
  • Failover rehearsed on the target platform at least annually.
  • Exit terms on both platforms, with portability evidence generated by operations.